CN115571656A - Automatic dumping control method and system based on material level detection - Google Patents

Automatic dumping control method and system based on material level detection Download PDF

Info

Publication number
CN115571656A
CN115571656A CN202211188584.0A CN202211188584A CN115571656A CN 115571656 A CN115571656 A CN 115571656A CN 202211188584 A CN202211188584 A CN 202211188584A CN 115571656 A CN115571656 A CN 115571656A
Authority
CN
China
Prior art keywords
matrix
level
decoding
training
material level
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202211188584.0A
Other languages
Chinese (zh)
Other versions
CN115571656B (en
Inventor
曹鋆程
刘强
孙新佳
田�文明
沈洋
刘跃
房圆武
王志元
郑树坤
冯川
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Uaneng Yimin Coal Power Co Ltd
Original Assignee
Uaneng Yimin Coal Power Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Uaneng Yimin Coal Power Co Ltd filed Critical Uaneng Yimin Coal Power Co Ltd
Priority to CN202211188584.0A priority Critical patent/CN115571656B/en
Publication of CN115571656A publication Critical patent/CN115571656A/en
Application granted granted Critical
Publication of CN115571656B publication Critical patent/CN115571656B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • BPERFORMING OPERATIONS; TRANSPORTING
    • B65CONVEYING; PACKING; STORING; HANDLING THIN OR FILAMENTARY MATERIAL
    • B65GTRANSPORT OR STORAGE DEVICES, e.g. CONVEYORS FOR LOADING OR TIPPING, SHOP CONVEYOR SYSTEMS OR PNEUMATIC TUBE CONVEYORS
    • B65G65/00Loading or unloading
    • B65G65/005Control arrangements
    • BPERFORMING OPERATIONS; TRANSPORTING
    • B65CONVEYING; PACKING; STORING; HANDLING THIN OR FILAMENTARY MATERIAL
    • B65GTRANSPORT OR STORAGE DEVICES, e.g. CONVEYORS FOR LOADING OR TIPPING, SHOP CONVEYOR SYSTEMS OR PNEUMATIC TUBE CONVEYORS
    • B65G65/00Loading or unloading
    • B65G65/28Piling or unpiling loose materials in bulk, e.g. coal, manure, timber, not otherwise provided for
    • GPHYSICS
    • G01MEASURING; TESTING
    • G01FMEASURING VOLUME, VOLUME FLOW, MASS FLOW OR LIQUID LEVEL; METERING BY VOLUME
    • G01F23/00Indicating or measuring liquid level or level of fluent solid material, e.g. indicating in terms of volume or indicating by means of an alarm
    • G01F23/22Indicating or measuring liquid level or level of fluent solid material, e.g. indicating in terms of volume or indicating by means of an alarm by measuring physical variables, other than linear dimensions, pressure or weight, dependent on the level to be measured, e.g. by difference of heat transfer of steam or water
    • G01F23/28Indicating or measuring liquid level or level of fluent solid material, e.g. indicating in terms of volume or indicating by means of an alarm by measuring physical variables, other than linear dimensions, pressure or weight, dependent on the level to be measured, e.g. by difference of heat transfer of steam or water by measuring the variations of parameters of electromagnetic or acoustic waves applied directly to the liquid or fluent solid material
    • G01F23/284Electromagnetic waves
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/08Learning methods
    • G06N3/084Backpropagation, e.g. using gradient descent
    • BPERFORMING OPERATIONS; TRANSPORTING
    • B65CONVEYING; PACKING; STORING; HANDLING THIN OR FILAMENTARY MATERIAL
    • B65GTRANSPORT OR STORAGE DEVICES, e.g. CONVEYORS FOR LOADING OR TIPPING, SHOP CONVEYOR SYSTEMS OR PNEUMATIC TUBE CONVEYORS
    • B65G2201/00Indexing codes relating to handling devices, e.g. conveyors, characterised by the type of product or load being conveyed or handled
    • B65G2201/04Bulk
    • B65G2201/045Sand, soil and mineral ore
    • YGENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02ATECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A40/00Adaptation technologies in agriculture, forestry, livestock or agroalimentary production
    • Y02A40/10Adaptation technologies in agriculture, forestry, livestock or agroalimentary production in agriculture

Landscapes

  • Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mechanical Engineering (AREA)
  • Electromagnetism (AREA)
  • Theoretical Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Biophysics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Thermal Sciences (AREA)
  • Fluid Mechanics (AREA)
  • Radar Systems Or Details Thereof (AREA)

Abstract

The application discloses automatic dumping control method and system based on material level detection, it adopts the level value jointly through a plurality of radar charge level indicators that arrange with the area array form to can take the diagonal value in the dumping area into consideration based on the relevance characteristic between each level value, with the received signal intensity relevance characteristic of a plurality of radar charge level indicators when measuring shows the measurement accuracy influence that signal interference brought, fuses these two implicit characteristic information and decodes the regression and obtains more accurately the correction measured value of material level. Thus, whether to perform soil discharge is determined based on the corrected measurement value, thereby improving the accuracy of automatic soil discharge control.

Description

Automatic dumping control method and system based on material level detection
Technical Field
The present application relates to the field of a dumping technique, and more particularly, to an automatic dumping control method and system based on material level detection.
Technical Field
Radar level gauges are measuring instruments based on the time-travel principle, the radar wave runs at the speed of light, the running time can be converted into a level signal through electronic components, a probe emits a high-frequency pulse and propagates along a cable, the pulse is reflected back when encountering the surface of a material and is received by a receiver in the instrument, and a distance signal is converted into a level signal.
Electromagnetic waves can penetrate through space steam, dust and other interference sources, and can be easily reflected when meeting obstacles, the better the conductivity of a measured medium or the higher the dielectric constant, the better the reflection effect of echo signals.
Transmission-reflection-reception is the basic principle of the operation of radar level gauges. The antenna of the radar sensor emits a radar signal of a minimum of 5.8GHz in the form of a beam. The reflected signal is still received by the antenna, and the time of travel of the radar level gauge radar pulse signal from transmission to reception is proportional to the distance of the sensor to the surface of the medium and the level.
At present, the level height is detected by installing a radar level gauge on the existing dumping machine so as to realize automatic dumping. However, when measuring solid materials, the sensor is tilted at a certain angle due to the solid media having a stack angle. Accordingly, when the radar level gauge is used for level measurement, the pile angle of the soil discharge belt changes all the time, so that the measurement deviation of the radar level gauge occurs, and further the control deviation occurs.
Therefore, an optimized automatic dumping control scheme based on level detection is desired.
Disclosure of Invention
The present application is proposed to solve the above-mentioned technical problems. The embodiment of the application provides an automatic dumping control method based on level detection and a system thereof, which jointly adopt the level value through a plurality of radar level gauges arranged in an area array form, so that the diagonal value of a dumping belt can be taken into consideration based on the relevance characteristics between the level values, the measurement precision influence brought by signal interference is represented by the received signal strength relevance characteristics of the radar level gauges during measurement, and the implicit characteristic information of the radar level gauges and the radar level gauges is fused to decode and regress to obtain a more accurate correction measurement value of the level. Thus, whether to perform soil discharge is determined based on the corrected measurement value, thereby improving the accuracy of automatic soil discharge control.
According to an aspect of the present application, there is provided an automatic soil discharge control method based on level detection, including:
acquiring a plurality of level values collected by a plurality of radar level gauges arranged in an area array form;
arranging the plurality of level values into a level input matrix according to the area array form;
obtaining a material level characteristic matrix by the trained first convolutional neural network model using a space attention mechanism through the material level input matrix;
obtaining received signal strength values of the plurality of radar level gauges when measuring level values;
arranging the received signal strength values of the plurality of radar level gauges when measuring level values into a signal strength input matrix in the form of the area array;
obtaining a signal intensity characteristic matrix by the trained second convolution neural network model using a space attention mechanism;
fusing the signal intensity characteristic matrix and the material level characteristic matrix to obtain a decoding characteristic matrix;
performing decoding regression on the decoding characteristic matrix through a trained decoder to obtain a decoding value for representing the level correction measured value; and
and determining whether to perform dumping based on the decoded value.
In the above automatic dumping control method based on material level detection, obtaining a material level feature matrix by training the material level input matrix through a first convolution neural network model using a space attention mechanism includes: depth convolution encoding the fill level input matrix using a convolution encoding portion of the first convolutional neural network model to obtain an initial convolution signature; inputting the initial convolution feature map into a spatial attention portion of the first convolutional neural network model to obtain a first spatial attention map; passing the first spatial attention map through a Softmax activation function to obtain a first spatial attention feature map; calculating the multiplication of the first spatial attention feature map and the initial convolution feature map according to position points to obtain a bin level feature map; and carrying out global mean pooling along channel dimensions on the material level characteristic diagram to obtain the material level characteristic matrix.
In the above automatic dumping control method based on material level detection, the passing the signal intensity input matrix through a trained second convolutional neural network model using a spatial attention mechanism to obtain a signal intensity characteristic matrix includes: performing deep convolutional encoding on the signal intensity input matrix by using a convolutional encoding part of the second convolutional neural network model to obtain an initial convolutional characteristic diagram; inputting the initial convolution feature map into a spatial attention portion of the second convolutional neural network model to obtain a second spatial attention map; passing the second spatial attention map through a Softmax activation function to obtain a second spatial attention feature map; calculating the multiplication of the second spatial attention feature map and the initial convolution feature map according to position points to obtain a signal intensity feature map; and performing global mean pooling along channel dimensions on the signal strength characteristic diagram to obtain the signal strength characteristic matrix.
In the above automatic dumping control method based on material level detection, the fusing the signal intensity characteristic matrix and the material level characteristic matrix to obtain a decoding characteristic matrix includes: fusing the signal intensity characteristic matrix and the material level characteristic matrix according to the following formula to obtain a decoding characteristic matrix; wherein the formula is:
Figure 318612DEST_PATH_IMAGE002
wherein ,
Figure DEST_PATH_IMAGE003
for the purpose of said decoding of the feature matrix,
Figure 87985DEST_PATH_IMAGE004
for the matrix of signal strength characteristics in question,
Figure DEST_PATH_IMAGE005
is the material level characteristic matrix "
Figure 311024DEST_PATH_IMAGE006
"indicates the addition of elements at the corresponding positions of the signal strength characteristic matrix and the level characteristic matrix,
Figure DEST_PATH_IMAGE007
and
Figure 592970DEST_PATH_IMAGE008
is a weighting parameter for controlling a balance between the signal strength characteristic matrix and the level characteristic matrix in the decoded characteristic matrix.
In the automatic dumping control method based on material level detection, the decoding characteristic matrix is subjected to decoding regression through a trained decoder to obtain a decoding value used for representing a material level correction measured valueValues, including: performing decoding regression on the decoding feature matrix by using the decoder according to the following formula to obtain the decoding value; wherein the formula is:
Figure DEST_PATH_IMAGE009
, wherein
Figure 917772DEST_PATH_IMAGE010
Is the matrix of the decoded features of the image,
Figure DEST_PATH_IMAGE011
is the value of the said decoded value or values,
Figure 415137DEST_PATH_IMAGE012
is a matrix of weights that is a function of,
Figure DEST_PATH_IMAGE013
representing a matrix multiplication.
In the above automatic dumping control method based on the material level detection, determining whether to perform dumping based on the decoded value includes: determining whether to perform soil discharging based on a comparison between the decoded value and a predetermined threshold.
In the above automatic dumping control method based on the material level detection, further comprising: training the decoder, the first convolutional neural network model using a spatial attention mechanism, and the second convolutional neural network model using a spatial attention mechanism.
In the above automatic dumping control method based on material level detection, the training the decoder, the first convolutional neural network model using a spatial attention mechanism, and the second convolutional neural network model using a spatial attention mechanism includes: obtaining training data, the training data comprising: the system comprises a plurality of training level values acquired by the plurality of radar level gauges arranged in an area array form, training received signal strength values and real level values of the plurality of radar level gauges when the level values are measured; arranging the plurality of training material level values and the plurality of training received signal strength values into a training material level input matrix and a training signal strength input matrix according to the area array form; passing the training material level input matrix through the first convolution neural network model using the spatial attention mechanism to obtain a training material level characteristic matrix; passing the training signal intensity input matrix through the second convolutional neural network model using a spatial attention mechanism to obtain a training signal intensity feature matrix; fusing the training signal intensity characteristic matrix and the training material level characteristic matrix to obtain a training decoding characteristic matrix; passing the training decoded feature matrix through the decoder to obtain a decoding loss function value; and iteratively training the decoder, the first convolutional neural network model using the spatial attention mechanism, and the second convolutional neural network model using the spatial attention mechanism by a gradient descent back propagation algorithm based on the decoding loss function values, wherein in each iteration of training, a decoding feature vector obtained by unfolding the decoding feature matrix is iterated based on a weight matrix before and after updating of each iteration of the decoder.
In the above automatic dumping control method based on material level detection, in each iteration of training, based on the weight matrix of the decoder before and after each iteration update, the decoding feature vector obtained by expanding the decoding feature matrix is iterated by the following formula, where the formula is:
Figure DEST_PATH_IMAGE015
wherein
Figure 695946DEST_PATH_IMAGE016
Representing the decoding feature vector obtained after the decoding feature matrix is expanded,
Figure DEST_PATH_IMAGE017
and
Figure 696263DEST_PATH_IMAGE018
respectively representing the weights of the decoder before and after each iteration updateThe matrix is a matrix of a plurality of matrices,
Figure DEST_PATH_IMAGE019
which represents the zero norm of the vector,
Figure 492049DEST_PATH_IMAGE020
an exponential operation representing a vector representing a natural exponential function value raised to a power of a feature value at each position in the vector,
Figure 337646DEST_PATH_IMAGE006
indicating that the addition is by position,
Figure DEST_PATH_IMAGE021
it is meant a subtraction by position,
Figure 800857DEST_PATH_IMAGE013
representing a matrix multiplication.
According to another aspect of the present application, there is provided an automatic soil discharge control system based on material level detection, including:
a level value acquisition unit for acquiring a plurality of level values acquired by a plurality of radar level gauges arranged in an area array;
the material level value arrangement unit is used for arranging the material level values into a material level input matrix according to the area array form;
the material level value space attention unit is used for enabling the material level input matrix to pass through a trained first convolution neural network model using a space attention mechanism so as to obtain a material level characteristic matrix;
the signal intensity value acquisition unit is used for acquiring the received signal intensity values of the plurality of radar level gauges when measuring the level values;
the signal intensity arrangement unit is used for arranging the received signal intensity values of the plurality of radar level gauges in the form of the area array as a signal intensity input matrix when measuring the level values;
the signal intensity space attention unit is used for enabling the signal intensity input matrix to pass through a trained second convolutional neural network model using a space attention mechanism so as to obtain a signal intensity characteristic matrix;
the decoding characteristic matrix generating unit is used for fusing the signal intensity characteristic matrix and the material level characteristic matrix to obtain a decoding characteristic matrix;
the decoding regression unit is used for performing decoding regression on the decoding characteristic matrix through a trained decoder to obtain a decoding value for representing the level correction measured value; and
and the comparison unit is used for determining whether to carry out dumping or not based on the decoding value.
In the above automatic dumping control system based on level detection, the level value space attention unit is further configured to: performing deep convolutional coding on the material level input matrix by using a convolutional coding part of the first convolutional neural network model to obtain an initial convolutional characteristic diagram; inputting the initial convolution feature map into a spatial attention portion of the first convolutional neural network model to obtain a first spatial attention map; passing the first spatial attention map through a Softmax activation function to obtain a first spatial attention feature map; calculating the multiplication of the first spatial attention feature map and the initial convolution feature map according to the position points to obtain a material level feature map; and carrying out global mean pooling treatment along the channel dimension on the material level characteristic diagram to obtain the material level characteristic matrix.
In the above automatic dumping control system based on material level detection, the signal intensity space attention unit is further configured to: performing deep convolutional encoding on the signal intensity input matrix by using a convolutional encoding part of the second convolutional neural network model to obtain an initial convolutional characteristic diagram; inputting the initial convolution feature map into a spatial attention portion of the second convolutional neural network model to obtain a second spatial attention map; passing the second spatial attention map through a Softmax activation function to obtain a second spatial attention feature map; calculating the multiplication of the second spatial attention feature map and the initial convolution feature map according to the position points to obtain a signal intensity feature map; and performing global mean pooling along channel dimensions on the signal intensity characteristic diagram to obtain the signal intensity characteristic matrix.
In the above automatic dumping control system based on material level detection, the decoding feature matrix generating unit is further configured to: fusing the signal intensity characteristic matrix and the material level characteristic matrix to obtain a decoding characteristic matrix according to the following formula; wherein the formula is:
Figure 237654DEST_PATH_IMAGE002
wherein ,
Figure 252228DEST_PATH_IMAGE003
for the purpose of said decoding of the feature matrix,
Figure 167094DEST_PATH_IMAGE004
for the matrix of signal strength characteristics in question,
Figure 766703DEST_PATH_IMAGE005
is the material level characteristic matrix "
Figure 639981DEST_PATH_IMAGE006
"indicates the addition of elements at the corresponding positions of the signal strength characteristic matrix and the level characteristic matrix,
Figure 941518DEST_PATH_IMAGE007
and
Figure 660076DEST_PATH_IMAGE008
is a weighting parameter for controlling a balance between the signal strength characteristic matrix and the level characteristic matrix in the decoding characteristic matrix.
In the above automatic dumping control system based on material level detection, the decoding regression unit is further configured to: performing decoding regression on the decoding feature matrix by using the decoder according to the following formula to obtain the decoding value; wherein the formula is:
Figure 317453DEST_PATH_IMAGE009
, wherein
Figure 345321DEST_PATH_IMAGE010
Is the matrix of the decoded features of the image,
Figure 884887DEST_PATH_IMAGE011
is the value of the said decoded value or values,
Figure 672714DEST_PATH_IMAGE012
is a matrix of the weights that is,
Figure 981336DEST_PATH_IMAGE013
representing a matrix multiplication.
In the above automatic dumping control system based on material level detection, the comparing unit is further configured to: determining whether to perform soil discharging based on a comparison between the decoded value and a predetermined threshold.
In the above-mentioned automatic dumping control system based on material level detection, still include: a training module to train the decoder, the first convolutional neural network model using a spatial attention mechanism, and the second convolutional neural network model using a spatial attention mechanism. The training module that trains the decoder, the first convolutional neural network model using spatial attention, and the second convolutional neural network model using spatial attention, comprising: a training data acquisition unit configured to acquire training data, the training data including: the system comprises a plurality of training level values acquired by the plurality of radar level gauges arranged in an area array form, training received signal strength values and real level values of the plurality of radar level gauges when the level values are measured; the training data input matrix construction unit is used for arranging the training material level values and the training received signal strength values into a training material level input matrix and a training signal strength input matrix according to the area array form; the training material level characteristic matrix generating unit is used for enabling the training material level input matrix to pass through the first convolution neural network model using the space attention mechanism to obtain a training material level characteristic matrix; the training signal intensity characteristic matrix generating unit is used for enabling the training signal intensity input matrix to pass through the second convolutional neural network model using the space attention mechanism to obtain a training signal intensity characteristic matrix; the fusion unit is used for fusing the training signal intensity characteristic matrix and the training material level characteristic matrix to obtain a training decoding characteristic matrix; a decoding loss function value generating unit, configured to pass the training decoding feature matrix through the decoder to obtain a decoding loss function value; and an iterative training unit, configured to iteratively train the decoder, the first convolutional neural network model using the spatial attention mechanism, and the second convolutional neural network model using the spatial attention mechanism by using a gradient descent back propagation algorithm based on the decoding loss function value, wherein in each iteration of the training, a decoded feature vector expanded from the decoded feature matrix is iterated based on a weight matrix before and after updating of each iteration by the decoder.
In the above automatic dumping control system based on material level detection, the iterative training unit is further configured to: in each round of training iteration, based on the weight matrix of the decoder before and after each iteration update, iterating the decoded feature vector obtained by expanding the decoded feature matrix according to the following formula:
Figure 383367DEST_PATH_IMAGE022
wherein
Figure 941387DEST_PATH_IMAGE016
Representing the decoding feature vector obtained after the decoding feature matrix is expanded,
Figure 470589DEST_PATH_IMAGE017
and
Figure 885914DEST_PATH_IMAGE018
respectively representing the weight matrix of the decoder before and after each iteration of updating,
Figure 6317DEST_PATH_IMAGE019
which represents the zero-norm of the vector,
Figure 520475DEST_PATH_IMAGE020
an exponential operation representing a vector representing a natural exponential function value raised to a power of a feature value at each position in the vector,
Figure 384526DEST_PATH_IMAGE006
it is shown that the addition by position,
Figure 651428DEST_PATH_IMAGE021
it is meant a subtraction by position,
Figure 208311DEST_PATH_IMAGE013
representing a matrix multiplication.
Compared with the prior art, the automatic dumping control method and system based on the material level detection, which are provided by the application, adopt the aggregate level value jointly through the plurality of radar material level meters arranged in the form of the area array, so that the diagonal value of the dumping belt can be taken into consideration based on the relevance characteristics between the aggregate level values, the measurement precision influence brought by signal interference is represented by the received signal strength relevance characteristics of the plurality of radar material level meters during measurement, and the implicit characteristic information of the radar material level meters and the received signal strength relevance characteristics are fused to perform decoding regression to obtain a more accurate correction measurement value of the material level. Thus, whether to perform soil discharge is determined based on the corrected measurement value, thereby improving the accuracy of automatic soil discharge control.
Drawings
The above and other objects, features and advantages of the present application will become more apparent by describing in more detail embodiments of the present application with reference to the attached drawings. The accompanying drawings are included to provide a further understanding of the embodiments of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and together with the description serve to explain the principles of the application. In the drawings, like reference numbers generally indicate like parts or steps.
Fig. 1 illustrates an application scenario diagram of an automatic dumping control method based on level detection according to an embodiment of the present application.
Fig. 2 illustrates a flowchart of an automatic dumping control method based on level detection according to an embodiment of the present application.
Fig. 3 illustrates an architecture diagram of an automatic dumping control method based on level detection according to an embodiment of the present application.
Fig. 4 illustrates a flowchart for training the decoder, the first convolutional neural network model using the spatial attention mechanism, and the second convolutional neural network model using the spatial attention mechanism in the automatic dumping control method based on level detection according to an embodiment of the present application.
FIG. 5 illustrates a block diagram of an automatic dumping control system based on level detection according to an embodiment of the present application.
FIG. 6 illustrates a block diagram of a training module in an automatic dumping control system based on level detection according to an embodiment of the present application.
Detailed Description
Hereinafter, example embodiments according to the present application will be described in detail with reference to the accompanying drawings. It should be apparent that the described embodiments are only a few embodiments of the present application, and not all embodiments of the present application, and it should be understood that the present application is not limited to the example embodiments described herein.
Summary of the application
As described above, the conventional earth-moving machine is provided with a radar level gauge to detect the level of material, thereby realizing automatic earth moving. However, when measuring solid materials, the sensor is tilted at a certain angle due to the solid media having a stack angle. Accordingly, when the radar level gauge is used for level measurement, the pile angle of the soil discharge belt changes all the time, so that the measurement deviation of the radar level gauge occurs, and further the control deviation occurs. Therefore, an optimized automatic dumping control scheme based on level detection is desired.
Accordingly, in order to accurately perform material level correction measurement in real time, the accuracy of automatic dumping control is improved. Therefore, in the technical solution of the present application, it is desirable to simultaneously perform the measurement of the material level by installing a plurality of radar level gauges arranged in an area array using an artificial intelligence control technique based on deep learning. That is, the level values are collectively acquired by a plurality of radar level gauges arranged in an area array, so that the diagonal value of the soil discharging belt can be taken into consideration based on the correlation characteristics between the respective level values. In order to filter out signal interference generated among the plurality of radar level gauges and improve the measurement accuracy, the measurement accuracy influence caused by the signal interference is further represented by the received signal strength correlation characteristics of the plurality of radar level gauges during measurement. Then, the implicit characteristic information of the two types of the materials is fused to carry out decoding regression to obtain a more accurate correction measured value of the material level. Therefore, the material level can be accurately corrected and measured in real time, and the accuracy of automatic dumping control is improved.
Specifically, in the technical scheme of the application, firstly, a plurality of level values are collected through a plurality of radar level gauges arranged in an area array form, and the plurality of level values are arranged into a level input matrix according to the area array form. Therefore, the data information of the material level values which are acquired by the plurality of radar material level meters arranged in the area array mode and have spatial position information can be integrated, and the subsequent characteristic extraction is favorably improved. And then, carrying out feature mining on the material level input matrix through a convolution neural network model with excellent performance in the aspect of implicit associated feature extraction, wherein spatial associated features exist among the material level values in the material level input matrix with the area array form spatial feature information in the acquired arrangement of the material level values. Therefore, in order to take the diagonal values of the soil discharge zone into consideration in the spatial position correlation characteristic distribution information focused on the various level values, in the technical solution of the present application, a first convolution neural network model of a spatial attention mechanism is further used to process the level input matrix to extract spatial correlation implicit characteristics between the various level values, so as to obtain the level characteristic matrix.
Then, it is considered that when the level measurement is performed using the plurality of radar level gauges arranged in the area array, the measurement accuracy may be affected by signal interference generated between the respective radar level gauges. Therefore, in the technical solution of the present application, in order to filter out signal interference generated between the plurality of radar level gauges to improve the accuracy of measurement, the measurement accuracy influence caused by the signal interference is further represented by a received signal strength correlation matrix of the plurality of radar level gauges when measuring a level value. That is, specifically, the collected received signal strength values of the plurality of radar level gauges when measuring the level values are first arranged in the form of the area array as a signal strength input matrix. Then, because each received signal intensity value in the signal intensity input matrix arranged in the area array form also has characteristic distribution information with relevance in a spatial position, a second convolution neural network model of a spatial attention mechanism is further used for processing the signal intensity input matrix so as to extract implicit relevant characteristic distribution information of the received signal intensity value of each radar level gauge focusing on a spatial dimension, and therefore the measurement accuracy influence caused by the signal interference is represented, and the signal intensity characteristic matrix is obtained.
Further, the signal intensity characteristic matrix and the material level characteristic matrix can be fused for decoding regression to obtain a decoding value for representing the material level correction measured value. Like this, can be in the filtering signal interference between each radar charge level indicator with under the influence of the diagonal value in the soil discharge area, right the material level is accurately rectified and measured, and then improves the accuracy of soil discharge control.
In particular, in the technical solution of the present application, since the material level feature matrix includes a material level value cross-area array spatial position distribution correlation feature extracted by using a first convolutional neural network model of a spatial attention mechanism, and the signal intensity feature matrix includes a signal interference cross-area array spatial position distribution correlation feature extracted by using a second convolutional neural network model of the spatial attention mechanism, there are more local cross-area array spatial position distribution correlation patterns in a decoding feature matrix obtained by fusing the signal intensity feature matrix and the material level feature matrix, so that when the decoding feature matrix is subjected to decoding regression by a decoder, a weight matrix of the decoder is subjected to a heavier adaptation burden with the decoding feature matrix through parameter adjustment, which affects the training speed of the decoder and the accuracy of a decoding result.
Based on this, in the technical solution of the present application, in the training process of the decoder, a scene-dependent optimization of decoder iteration is introduced, specifically, a decoded feature vector obtained after the decoded feature matrix is expanded is used as the decoded feature vector
Figure 944186DEST_PATH_IMAGE016
Calculating its adapted optimized feature vector relative to the decoder weight matrix
Figure DEST_PATH_IMAGE023
Expressed as:
Figure 64457DEST_PATH_IMAGE024
Figure 467757DEST_PATH_IMAGE017
and
Figure 195542DEST_PATH_IMAGE018
is the weight matrix before and after each iteration update of the decoder,
Figure 153133DEST_PATH_IMAGE019
representing the zero norm of the vector.
That is, the decoded feature vectors are corrected by using the measure of the scene point correlation before and after the parameter update of the weight matrix at the time of the iteration of the decoder as a correction factor
Figure 873833DEST_PATH_IMAGE016
Is optimized to support the decoding feature vector by the distribution similarity of the decoding scenes of the decoder
Figure 866060DEST_PATH_IMAGE016
Performing correlation description to promote optimized decoding eigenvector while performing parameter update on weight matrix of decoder
Figure 499167DEST_PATH_IMAGE023
And the adaptive degree of the parameter updating is realized, so that the training speed of the decoder is accelerated, and the accuracy of the decoding result is improved. Therefore, the material level can be accurately corrected and measured in real time, and the accuracy of automatic dumping control is improved.
Based on this, the application provides an automatic dumping control method based on material level detection, which includes: acquiring a plurality of level values acquired by a plurality of radar level gauges arranged in an area array form; arranging the plurality of level values into a level input matrix according to the area array form; obtaining a material level characteristic matrix by the trained first convolutional neural network model using a space attention mechanism through the material level input matrix; obtaining received signal strength values of the plurality of radar level gauges when measuring level values; arranging the received signal strength values of the plurality of radar level gauges when measuring the level values into a signal strength input matrix in the form of the area array; obtaining a signal intensity characteristic matrix by the trained second convolution neural network model using a space attention mechanism through the signal intensity input matrix; fusing the signal intensity characteristic matrix and the material level characteristic matrix to obtain a decoding characteristic matrix; performing decoding regression on the decoding characteristic matrix through a trained decoder to obtain a decoding value for representing the level correction measured value; and determining whether to perform dumping based on the decoded value.
Fig. 1 illustrates an application scenario of an automatic dumping control method based on level detection according to an embodiment of the present application. As shown in fig. 1, in this application scenario, first, a plurality of level values collected by a plurality of radar level gauges (e.g., se1 to Sen as illustrated in fig. 1) arranged in an area array and received signal strength values of the plurality of radar level gauges when measuring the level values are obtained. Further, the plurality of level values and the plurality of received signal strength values are input into a server (e.g., S as illustrated in fig. 1) deployed with an automatic dumping control algorithm based on level detection, wherein the server is capable of processing the plurality of level values and the plurality of received signal strength values based on the automatic dumping control algorithm based on level detection to obtain decoded values representing level correction measurement values, and determining whether to perform dumping based on the decoded values.
Having described the general principles of the present application, various non-limiting embodiments of the present application will now be described with reference to the accompanying drawings.
Exemplary method
Fig. 2 illustrates a flowchart of an automatic dumping control method based on level detection according to an embodiment of the present application. As shown in fig. 2, the automatic dumping control method based on the material level detection according to the embodiment of the present application includes: s110, acquiring a plurality of level values acquired by a plurality of radar level gauges arranged in an area array form; s120, arranging the plurality of material level values into a material level input matrix according to the area array form; s130, obtaining a material level characteristic matrix by the material level input matrix through a trained first convolution neural network model using a space attention mechanism; s140, obtaining the received signal strength values of the plurality of radar level gauges when measuring the level values; s150, arranging the received signal strength values of the plurality of radar level gauges when measuring the level values into a signal strength input matrix in the form of the area array; s160, obtaining a signal intensity characteristic matrix by the signal intensity input matrix through a trained second convolutional neural network model using a space attention mechanism; s170, fusing the signal intensity characteristic matrix and the material level characteristic matrix to obtain a decoding characteristic matrix; s180, performing decoding regression on the decoding characteristic matrix through a trained decoder to obtain a decoding value for representing the measured material level correction value; and S190, determining whether to carry out dumping or not based on the decoding value.
Fig. 3 illustrates an architecture diagram of an automatic dumping control method based on level detection according to an embodiment of the present application. As shown in fig. 3, in the architecture diagram, a plurality of level values collected by a plurality of radar level gauges arranged in an area array form are first acquired. And then, arranging the plurality of level values into a level input matrix according to the area array form. Then, the material level input matrix passes through a trained first convolution neural network model using a spatial attention mechanism to obtain a material level characteristic matrix. And further, obtaining the received signal strength values of the plurality of radar level gauges when measuring the level values. Then, the received signal strength values of the plurality of radar level gauges when measuring the level values are arranged in the form of the area array as a signal strength input matrix. And then, passing the signal strength input matrix through a trained second convolution neural network model using a spatial attention mechanism to obtain a signal strength characteristic matrix. And then fusing the signal intensity characteristic matrix and the material level characteristic matrix to obtain a decoding characteristic matrix. Then, the decoding characteristic matrix is subjected to decoding regression through a trained decoder to obtain a decoding value for representing the material level correction measured value, and whether to perform dumping is determined based on the decoding value.
In step S110, a plurality of level values collected by a plurality of radar level gauges arranged in an area array are acquired. As described above, the conventional earth-moving machine is provided with a radar level gauge to detect the level of material, thereby realizing automatic earth moving. However, when measuring solid materials, the sensor is inclined at a certain angle because the solid medium has a pile angle. Accordingly, when the radar level gauge is used for level measurement, the pile angle of the soil discharge belt can change constantly, so that the measurement of the radar level gauge deviates, and further the control deviation is caused. Therefore, an optimized automatic dumping control scheme based on level detection is desired. Accordingly, in order to accurately perform material level correction measurement in real time, the accuracy of automatic dumping control is improved. Therefore, in the technical scheme of the application, the artificial intelligence control technology based on deep learning is expected to be used, and the level measurement is carried out simultaneously by arranging a plurality of radar level gauges in an area array mode. That is, the level values are collectively acquired by a plurality of radar level gauges arranged in an area array, so that the diagonal value of the soil discharging belt can be taken into consideration based on the correlation characteristics between the respective level values. In order to filter out signal interference generated among the plurality of radar level gauges and improve the measurement accuracy, the measurement accuracy influence caused by the signal interference is further represented by the received signal strength correlation characteristics of the plurality of radar level gauges during measurement. Then, the implicit characteristic information of the two types of the materials is fused to carry out decoding regression to obtain a more accurate correction measured value of the material level. Therefore, the material level can be accurately corrected and measured in real time, and the accuracy of automatic dumping control is improved. Specifically, in the technical scheme of the application, firstly, a plurality of level values are collected through a plurality of radar level gauges arranged in an area array form.
In step S120, the plurality of level values are arranged in the area array form as a level input matrix. Like this, can integrate the data information that has spatial position information the material level value that a plurality of radar charge level indicators arranged with the area array form gathered jointly is favorable to improving subsequent characteristic and draws.
In step S130, the material level input matrix is passed through a trained first convolution neural network model using a spatial attention mechanism to obtain a material level feature matrix. That is, the material level input matrix is subjected to feature mining through a convolution neural network model with excellent performance in implicit correlation feature extraction, but the spatial correlation feature is considered to exist among the material level values in the collected material level input matrix with the spatial feature information in the form of an area array in which the material level values are arranged. Therefore, in order to take the diagonal values of the soil discharge zone into consideration in the spatial position correlation characteristic distribution information focused on the various material level values, in the technical solution of the present application, a first convolutional neural network model of a spatial attention mechanism is further used to process the material level input matrix so as to extract spatial correlation implicit characteristics between the various material level values, thereby obtaining a material level characteristic matrix.
In one example, in the above automatic dumping control method based on material level detection, the passing the material level input matrix through a trained first convolutional neural network model using a spatial attention mechanism to obtain a material level feature matrix includes: performing deep convolutional coding on the material level input matrix by using a convolutional coding part of the first convolutional neural network model to obtain an initial convolutional characteristic diagram; inputting the initial convolution feature map into a spatial attention portion of the first convolutional neural network model to obtain a first spatial attention map; passing the first spatial attention map through a Softmax activation function to obtain a first spatial attention feature map; calculating the multiplication of the first spatial attention feature map and the initial convolution feature map according to the position points to obtain a material level feature map; and carrying out global mean pooling treatment along the channel dimension on the material level characteristic diagram to obtain the material level characteristic matrix.
In step S140, received signal strength values of the plurality of radar level gauges when measuring level values are obtained. It is considered that when level measurements are performed with the plurality of radar level gauges arranged in an area array, measurement accuracy may be affected due to signal interference generated between the respective radar level gauges. Therefore, in the technical solution of the present application, in order to filter out signal interference generated between the plurality of radar level gauges to improve the accuracy of measurement, the measurement accuracy influence caused by the signal interference is further represented by a received signal strength correlation matrix of the plurality of radar level gauges when measuring a level value. That is, the received signal strength values of the plurality of radar level gauges when measuring level values are first obtained.
In step S150, the received signal strength values of the plurality of radar level gauges when measuring level values are arranged in the form of the area array as a signal strength input matrix. That is, the collected received signal strength values of the plurality of radar level gauges when measuring the level values are arranged in the form of the area array as a signal strength input matrix.
In step S160, the signal strength input matrix is passed through the trained second convolutional neural network model using the spatial attention mechanism to obtain a signal strength feature matrix. Since each received signal intensity value in the signal intensity input matrix arranged in the area array form also has associated feature distribution information at a spatial position, the signal intensity input matrix is further processed by using a second convolutional neural network model of a spatial attention mechanism to extract implicit associated feature distribution information of the received signal intensity value of each radar level gauge focusing on a spatial dimension, so as to represent the measurement accuracy influence caused by the signal interference, and thus a signal intensity feature matrix is obtained.
In one example, in the above automatic dumping control method based on material level detection, the passing the signal strength input matrix through a trained second convolutional neural network model using a spatial attention mechanism to obtain a signal strength characteristic matrix includes: depth convolution encoding the signal strength input matrix using a convolution encoding portion of the second convolutional neural network model to obtain an initial convolution feature map; inputting the initial convolution feature map into a spatial attention portion of the second convolutional neural network model to obtain a second spatial attention map; passing the second spatial attention map through a Softmax activation function to obtain a second spatial attention feature map; calculating the multiplication of the second spatial attention feature map and the initial convolution feature map according to position points to obtain a signal intensity feature map; and performing global mean pooling along channel dimensions on the signal intensity characteristic diagram to obtain the signal intensity characteristic matrix.
In step S170, the signal strength characteristic matrix and the material level characteristic matrix are fused to obtain a decoding characteristic matrix. Further, the signal strength characteristic matrix and the material level characteristic matrix can be fused for decoding regression.
In one example, in the above automatic dumping control method based on material level detection, the fusing the signal intensity feature matrix and the material level feature matrix to obtain a decoded feature matrix includes: fusing the signal intensity characteristic matrix and the material level characteristic matrix according to the following formula to obtain a decoding characteristic matrix; wherein the formula is:
Figure 659234DEST_PATH_IMAGE002
wherein ,
Figure 934358DEST_PATH_IMAGE003
for the purpose of said decoding of the feature matrix,
Figure 515512DEST_PATH_IMAGE004
for the matrix of signal strength characteristics in question,
Figure 585099DEST_PATH_IMAGE005
is the material level characteristic matrix "
Figure 32130DEST_PATH_IMAGE006
"indicates the addition of elements at the corresponding positions of the signal strength characteristic matrix and the level characteristic matrix,
Figure 845365DEST_PATH_IMAGE007
and
Figure 812184DEST_PATH_IMAGE008
is a weighting parameter for controlling a balance between the signal strength characteristic matrix and the level characteristic matrix in the decoding characteristic matrix.
In step S180, the decoding characteristic matrix is subjected to decoding regression through a trained decoder to obtain a decoded value representing the level correction measurement value. The decoded values filter out signal interference between the radar level gauges and the influence of the diagonal values of the soil discharge belt.
In one example, the base material is as described aboveIn the method for controlling automatic dumping through bit detection, the decoding regression of the decoding characteristic matrix through a trained decoder to obtain a decoding value for representing a correction measurement value of the material level comprises the following steps: performing decoding regression on the decoding feature matrix by using the decoder according to the following formula to obtain the decoding value; wherein the formula is:
Figure 787093DEST_PATH_IMAGE009
, wherein
Figure 924682DEST_PATH_IMAGE010
Is the matrix of the decoded features of the image,
Figure 276029DEST_PATH_IMAGE011
is the value of the said decoded value or values,
Figure 97355DEST_PATH_IMAGE012
is a matrix of the weights that is,
Figure 492433DEST_PATH_IMAGE013
representing a matrix multiplication.
In step S190, it is determined whether to perform soil discharging based on the decoded value. Like this, can be in the filtering signal interference between each radar charge level indicator with under the influence of the diagonal value in the soil discharge area, right the charge level is accurately rectified and is measured, and then improves the accuracy of soil discharge control.
In one example, in the above automatic soil discharging control method based on level detection, the determining whether to discharge soil based on the decoded value includes: determining whether to perform soil discharging based on a comparison between the decoded value and a predetermined threshold.
In the above automatic dumping control method based on the material level detection, further comprising: training the decoder, the first convolutional neural network model using a spatial attention mechanism, and the second convolutional neural network model using a spatial attention mechanism.
Fig. 4 illustrates a flowchart for training the decoder, the first convolutional neural network model using the spatial attention mechanism, and the second convolutional neural network model using the spatial attention mechanism in the automatic dumping control method based on level detection according to an embodiment of the present application. As shown in fig. 4, in the above automatic dumping control method based on material level detection, the training of the decoder, the first convolutional neural network model using spatial attention mechanism and the second convolutional neural network model using spatial attention mechanism includes: s210, obtaining training data, wherein the training data comprises: the system comprises a plurality of training level values acquired by the plurality of radar level gauges arranged in an area array form, training received signal strength values and real level values of the plurality of radar level gauges when the level values are measured; s220, arranging the training material level values and the training received signal strength values into a training material level input matrix and a training signal strength input matrix according to the area array form; s230, enabling the training material level input matrix to pass through the first convolution neural network model using the space attention mechanism to obtain a training material level characteristic matrix; s240, enabling the training signal intensity input matrix to pass through the second convolutional neural network model using the spatial attention mechanism to obtain a training signal intensity characteristic matrix; s250, fusing the training signal intensity characteristic matrix and the training material level characteristic matrix to obtain a training decoding characteristic matrix; s260, enabling the training decoding characteristic matrix to pass through the decoder to obtain a decoding loss function value; and S270, iteratively training the decoder, the first convolutional neural network model using the spatial attention mechanism, and the second convolutional neural network model using the spatial attention mechanism by a gradient descent back-propagation algorithm based on the decoding loss function values, wherein in each iteration of training, the decoding feature vector expanded by the decoding feature matrix is iterated based on the weight matrix before and after updating of each iteration by the decoder.
In particular, in the technical solution of the present application, since the material level feature matrix includes a material level value cross-area array spatial position distribution correlation feature extracted by using a first convolutional neural network model of a spatial attention mechanism, and the signal intensity feature matrix includes a signal interference cross-area array spatial position distribution correlation feature extracted by using a second convolutional neural network model of the spatial attention mechanism, there are more local cross-area array spatial position distribution correlation patterns in a decoding feature matrix obtained by fusing the signal intensity feature matrix and the material level feature matrix, so that when the decoding feature matrix is subjected to decoding regression by a decoder, a weight matrix of the decoder is subjected to a heavier adaptation burden with the decoding feature matrix through parameter adjustment, which affects the training speed of the decoder and the accuracy of a decoding result. Based on this, in the technical solution of the present application, in the training process of the decoder, a scene dependent optimization of decoder iteration is introduced.
In one example, in the above automatic dumping control method based on material level detection, in each iteration of training, based on the weight matrix of the decoder before and after updating in each iteration, the decoded feature vector obtained by expanding the decoded feature matrix is iterated according to the following formula:
Figure 664788DEST_PATH_IMAGE024
wherein
Figure 554247DEST_PATH_IMAGE016
Representing the decoding feature vector obtained after the decoding feature matrix is expanded,
Figure 495658DEST_PATH_IMAGE017
and
Figure 330146DEST_PATH_IMAGE018
respectively representing the weight matrix of the decoder before and after each iteration of updating,
Figure 724218DEST_PATH_IMAGE019
which represents the zero norm of the vector,
Figure 417368DEST_PATH_IMAGE020
an exponential operation representing a vector representing a natural exponential function value raised to a power of a feature value at each position in the vector,
Figure 213286DEST_PATH_IMAGE006
it is shown that the addition by position,
Figure 215746DEST_PATH_IMAGE021
it is meant a subtraction by position,
Figure 34797DEST_PATH_IMAGE013
representing a matrix multiplication.
That is, the decoded feature vectors are corrected by using the measure of the scene point correlation before and after the parameter update of the weight matrix at the time of the iteration of the decoder as a correction factor
Figure 797217DEST_PATH_IMAGE016
Is optimized to support the decoding feature vector by the distribution similarity of the decoding scenes of the decoder
Figure 900171DEST_PATH_IMAGE016
Performing correlation description to promote optimized decoding eigenvector while performing parameter update on weight matrix of decoder
Figure 824264DEST_PATH_IMAGE023
And the adaptive degree of the parameter updating is realized, so that the training speed of the decoder is accelerated, and the accuracy of the decoding result is improved. Therefore, the material level can be accurately corrected and measured in real time, and the accuracy of automatic dumping control is improved.
In summary, the method for controlling automatic soil discharging based on level detection according to the embodiment of the present application is clarified, and the level values are collected by a plurality of radar level gauges arranged in an area array manner, so that the diagonal values of the soil discharging belt can be taken into consideration based on the correlation characteristics between the level values, the measurement accuracy influence caused by signal interference is represented by the correlation characteristics of the received signal intensities of the radar level gauges during measurement, and the implicit characteristic information of the radar level gauges and the received signal intensities is fused to perform decoding regression to obtain a more accurate corrected measurement value of the level. Thus, whether to perform soil discharging is determined based on the corrected measurement value, thereby improving the accuracy of automatic soil discharging control.
Exemplary System
FIG. 5 illustrates a block diagram of an automatic dumping control system based on level detection according to an embodiment of the present application. As shown in fig. 5, the automatic soil discharge control system 100 based on level detection according to the embodiment of the present application includes: a level value acquiring unit 110 for acquiring a plurality of level values collected by a plurality of radar level gauges arranged in an area array; a level value arranging unit 120, configured to arrange the plurality of level values into a level input matrix according to the area array form; a level value space attention unit 130, configured to obtain a level feature matrix by passing the level input matrix through a trained first convolution neural network model using a space attention mechanism; a signal strength value obtaining unit 140 for obtaining received signal strength values of the plurality of radar level gauges when measuring level values; the signal strength arrangement unit 150 is used for arranging the received signal strength values of the plurality of radar level gauges during measuring the level values into a signal strength input matrix in the form of the area array; a signal strength spatial attention unit 160, configured to obtain a signal strength feature matrix by passing the signal strength input matrix through a trained second convolutional neural network model using a spatial attention mechanism; a decoding characteristic matrix generating unit 170, configured to fuse the signal strength characteristic matrix and the material level characteristic matrix to obtain a decoding characteristic matrix; a decoding regression unit 180, configured to perform decoding regression on the decoding feature matrix through a trained decoder to obtain a decoded value representing the level correction measurement value; and a comparing unit 190 for determining whether to perform soil discharging based on the decoded value.
In one example, in the above automatic soil discharging control system 100 based on material level detection, the material level value space attention unit 130 is further configured to: performing deep convolutional coding on the material level input matrix by using a convolutional coding part of the first convolutional neural network model to obtain an initial convolutional characteristic diagram; inputting the initial convolution feature map into a spatial attention portion of the first convolutional neural network model to obtain a first spatial attention map; passing the first spatial attention map through a Softmax activation function to obtain a first spatial attention feature map; calculating the multiplication of the first spatial attention feature map and the initial convolution feature map according to the position points to obtain a material level feature map; and carrying out global mean pooling along channel dimensions on the material level characteristic diagram to obtain the material level characteristic matrix.
In one example, in the above automatic dumping control system 100 based on level detection, the signal strength space attention unit 160 is further configured to: depth convolution encoding the signal strength input matrix using a convolution encoding portion of the second convolutional neural network model to obtain an initial convolution feature map; inputting the initial convolution feature map into a spatial attention portion of the second convolutional neural network model to obtain a second spatial attention map; the second space attention diagram is activated through a Softmax activation function to obtain a second space attention feature diagram; calculating the multiplication of the second spatial attention feature map and the initial convolution feature map according to the position points to obtain a signal intensity feature map; and performing global mean pooling along channel dimensions on the signal intensity characteristic diagram to obtain the signal intensity characteristic matrix.
In an example, in the above automatic dumping control system 100 based on level detection, the decoding feature matrix generating unit 170 is further configured to: fusing the signal intensity characteristic matrix and the material level characteristic matrix to obtain a decoding characteristic matrix according to the following formula; wherein the formula is:
Figure 458508DEST_PATH_IMAGE002
wherein ,
Figure 477149DEST_PATH_IMAGE003
for the purpose of said decoding of the feature matrix,
Figure 982079DEST_PATH_IMAGE004
is a matrix of the signal strength characteristics for the signal,
Figure 77074DEST_PATH_IMAGE005
is a characteristic matrix of the material level'
Figure 667455DEST_PATH_IMAGE006
"means that the elements at the corresponding positions of the signal strength characteristic matrix and the material level characteristic matrix are added,
Figure 18016DEST_PATH_IMAGE007
and
Figure 377453DEST_PATH_IMAGE008
is a weighting parameter for controlling a balance between the signal strength characteristic matrix and the level characteristic matrix in the decoding characteristic matrix.
In one example, in the above automatic dumping control system 100 based on level detection, the decoding regression unit 180 is further configured to: performing decoding regression on the decoding feature matrix by using the decoder according to the following formula to obtain the decoding value; wherein the formula is:
Figure 643349DEST_PATH_IMAGE009
, wherein
Figure 721026DEST_PATH_IMAGE010
Is the matrix of the decoded features of the image,
Figure 347049DEST_PATH_IMAGE011
is the value of the said decoded value or values,
Figure 826572DEST_PATH_IMAGE012
is a matrix of weights that is a function of,
Figure 263369DEST_PATH_IMAGE013
representing a matrix multiplication.
In one example, in the automatic soil discharging control system 100 based on material level detection, the comparing unit 190 is further configured to: determining whether to perform soil discharging based on a comparison between the decoded value and a predetermined threshold.
In the above automatic soil discharge control system 100 based on material level detection, the system further includes: a training module 200 that trains the decoder, the first convolutional neural network model using spatial attention, and the second convolutional neural network model using spatial attention.
FIG. 6 illustrates a block diagram of a training module in an automatic dumping control system based on level detection according to an embodiment of the present application. As shown in fig. 6, the training module 200 for training the decoder, the first convolutional neural network model using the spatial attention mechanism, and the second convolutional neural network model using the spatial attention mechanism includes: a training data obtaining unit 210, configured to obtain training data, where the training data includes: a plurality of training level values acquired by the plurality of radar level gauges arranged in an area array form, training received signal strength values and real level values of the plurality of radar level gauges when the level values are measured; a training data input matrix constructing unit 220, configured to arrange the multiple training level values and the multiple training received signal strength values into a training level input matrix and a training signal strength input matrix according to the area array form; a training level feature matrix generating unit 230, configured to pass the training level input matrix through the first convolutional neural network model using the spatial attention mechanism to obtain a training level feature matrix; a training signal intensity feature matrix generating unit 240, configured to pass the training signal intensity input matrix through the second convolutional neural network model using the spatial attention mechanism to obtain a training signal intensity feature matrix; a fusion unit 250, configured to fuse the training signal intensity feature matrix and the training material level feature matrix to obtain a training decoding feature matrix; a decoding loss function value generating unit 260, configured to pass the training decoding feature matrix through the decoder to obtain a decoding loss function value; and an iterative training unit 270, configured to iteratively train the decoder, the first convolutional neural network model using the spatial attention mechanism, and the second convolutional neural network model using the spatial attention mechanism by using a gradient descent back propagation algorithm based on the decoding loss function value, where in each iteration of the training, a decoded feature vector expanded from the decoded feature matrix is iterated based on a weight matrix before and after updating of each iteration by the decoder.
In one example, in the automatic dumping control system 100 based on level detection, the iterative training unit 270 is further configured to: in each iteration of training, based on the weight matrix of the decoder before and after each iteration update, iterating the decoded feature vector obtained by expanding the decoded feature matrix according to the following formula:
Figure 828343DEST_PATH_IMAGE022
wherein
Figure 523635DEST_PATH_IMAGE016
Representing the decoding feature vector obtained after the decoding feature matrix is expanded,
Figure 123244DEST_PATH_IMAGE017
and
Figure 730943DEST_PATH_IMAGE018
respectively representing the weight matrix of the decoder before and after each iteration update,
Figure 970163DEST_PATH_IMAGE019
which represents the zero norm of the vector,
Figure 954299DEST_PATH_IMAGE020
an exponential operation representing a vector representing a natural exponential function value raised to a power of a feature value at each position in the vector,
Figure 408415DEST_PATH_IMAGE006
indicating that the addition is by position,
Figure 187015DEST_PATH_IMAGE021
it is meant a subtraction by position,
Figure 713198DEST_PATH_IMAGE013
representing a matrix multiplication.
Here, it may be understood by those skilled in the art that the specific functions and operations of the respective units and modules in the above-described automatic soil discharge control system 100 based on level detection have been described in detail in the above description of the automatic soil discharge control method based on level detection with reference to fig. 1 to 4, and thus, a repetitive description thereof will be omitted.
As described above, the automatic dumping control system 100 based on level detection according to the embodiment of the present application may be implemented in various terminal devices, such as a server for automatic dumping control based on level detection, and the like. In one example, the automatic dumping control system 100 based on level detection according to the embodiment of the present application may be integrated into a terminal device as one software module and/or hardware module. For example, the automatic dumping control system based on level detection 100 may be a software module in the operating system of the terminal device or may be an application developed for the terminal device; of course, the automatic dumping control system 100 based on level detection can also be one of numerous hardware modules of the terminal device.
Alternatively, in another example, the automatic soil discharging control system based on level detection 100 and the terminal device may also be separate devices, and the automatic soil discharging control system based on level detection 100 may be connected to the terminal device through a wired and/or wireless network and transmit the interactive information according to an agreed data format.

Claims (10)

1. An automatic dumping control method based on material level detection is characterized by comprising the following steps:
acquiring a plurality of level values acquired by a plurality of radar level gauges arranged in an area array form;
arranging the plurality of level values into a level input matrix according to the area array form;
obtaining a material level characteristic matrix by the trained first convolutional neural network model using a space attention mechanism through the material level input matrix;
obtaining received signal strength values of the plurality of radar level gauges when measuring level values;
arranging the received signal strength values of the plurality of radar level gauges when measuring level values into a signal strength input matrix in the form of the area array;
obtaining a signal intensity characteristic matrix by the trained second convolution neural network model using a space attention mechanism;
fusing the signal intensity characteristic matrix and the material level characteristic matrix to obtain a decoding characteristic matrix;
performing decoding regression on the decoding characteristic matrix through a trained decoder to obtain a decoding value for representing the level correction measurement value; and
and determining whether to perform dumping based on the decoded value.
2. The automatic dumping control method based on material level detection as claimed in claim 1, wherein said passing said material level input matrix through a trained first convolution neural network model using spatial attention mechanism to obtain a material level feature matrix comprises:
depth convolution encoding the fill level input matrix using a convolution encoding portion of the first convolutional neural network model to obtain an initial convolution signature;
inputting the initial convolution feature map into a spatial attention portion of the first convolutional neural network model to obtain a first spatial attention map;
the first space attention diagram is activated through a Softmax activation function to obtain a first space attention feature diagram;
calculating the multiplication of the first spatial attention feature map and the initial convolution feature map according to the position points to obtain a material level feature map; and
and carrying out global mean pooling treatment along the channel dimension on the material level characteristic diagram to obtain the material level characteristic matrix.
3. The method as claimed in claim 2, wherein the passing the signal intensity input matrix through a trained second convolutional neural network model using a spatial attention mechanism to obtain a signal intensity feature matrix comprises:
depth convolution encoding the signal strength input matrix using a convolution encoding portion of the second convolutional neural network model to obtain an initial convolution feature map;
inputting the initial convolution feature map into a spatial attention portion of the second convolutional neural network model to obtain a second spatial attention map;
passing the second spatial attention map through a Softmax activation function to obtain a second spatial attention feature map;
calculating the multiplication of the second spatial attention feature map and the initial convolution feature map according to the position points to obtain a signal intensity feature map; and
and carrying out global mean pooling processing along the channel dimension on the signal intensity characteristic diagram to obtain the signal intensity characteristic matrix.
4. The method as claimed in claim 3, wherein the fusing the signal strength characteristic matrix and the material level characteristic matrix to obtain a decoding characteristic matrix comprises: fusing the signal intensity characteristic matrix and the material level characteristic matrix to obtain a decoding characteristic matrix according to the following formula;
wherein the formula is:
Figure 284305DEST_PATH_IMAGE001
wherein ,
Figure 858506DEST_PATH_IMAGE002
for the purpose of said decoding of the feature matrix,
Figure 808008DEST_PATH_IMAGE003
is a matrix of the signal strength characteristics for the signal,
Figure 287399DEST_PATH_IMAGE004
is the material level characteristic matrix "
Figure 613338DEST_PATH_IMAGE005
"means that the elements at the corresponding positions of the signal strength characteristic matrix and the material level characteristic matrix are added,
Figure 245308DEST_PATH_IMAGE006
and
Figure 880558DEST_PATH_IMAGE007
is a weighting parameter for controlling a balance between the signal strength characteristic matrix and the level characteristic matrix in the decoding characteristic matrix.
5. The method as claimed in claim 4, wherein the decoding and regressing the decoding feature matrix through a trained decoder to obtain a decoded value representing a level correction measured value comprises:
performing decoding regression on the decoding feature matrix by using the decoder according to the following formula to obtain the decoding value; wherein the formula is:
Figure 332399DEST_PATH_IMAGE008
, wherein
Figure 183068DEST_PATH_IMAGE009
Is the matrix of the decoded features of the image,
Figure 669544DEST_PATH_IMAGE010
is the value of the said decoded value or values,
Figure 226427DEST_PATH_IMAGE011
is a matrix of the weights that is,
Figure 211569DEST_PATH_IMAGE012
representing a matrix multiplication.
6. The automatic dumping control method based on the level detection as claimed in claim 5, wherein said determining whether to perform dumping based on said decoded value includes:
determining whether to perform soil discharging based on a comparison between the decoded value and a predetermined threshold.
7. The automatic dumping control method based on the level detection according to claim 1, further comprising: training the decoder, the first convolutional neural network model using a spatial attention mechanism, and the second convolutional neural network model using a spatial attention mechanism.
8. The method of claim 7, wherein training the decoder, the first convolutional neural network model using a spatial attention mechanism, and the second convolutional neural network model using a spatial attention mechanism comprises:
obtaining training data, the training data comprising: the system comprises a plurality of training level values acquired by the plurality of radar level gauges arranged in an area array form, training received signal strength values and real level values of the plurality of radar level gauges when the level values are measured;
arranging the training material level values and the training receiving signal strength values into a training material level input matrix and a training signal strength input matrix according to the area array form;
passing the training material level input matrix through the first convolution neural network model using the spatial attention mechanism to obtain a training material level characteristic matrix;
passing the training signal intensity input matrix through the second convolutional neural network model using a spatial attention mechanism to obtain a training signal intensity feature matrix;
fusing the training signal intensity characteristic matrix and the training material level characteristic matrix to obtain a training decoding characteristic matrix;
passing the training decoded feature matrix through the decoder to obtain a decoding loss function value; and
iteratively training the decoder, the first convolutional neural network model using the spatial attention mechanism, and the second convolutional neural network model using the spatial attention mechanism based on the decoding loss function values and with a gradient descent back propagation algorithm, wherein in each iteration of training, the decoding feature vector expanded by the decoding feature matrix is iterated based on the weight matrix of the decoder before and after updating in each iteration.
9. The method of claim 8, wherein in each iteration of the training, based on the weight matrix of the decoder before and after each iteration, the decoded feature vector obtained by the expansion of the decoded feature matrix is iterated according to the following formula:
Figure 82573DEST_PATH_IMAGE013
wherein
Figure 220294DEST_PATH_IMAGE014
Representing the decoding feature vector obtained after the decoding feature matrix is expanded,
Figure 931767DEST_PATH_IMAGE015
and
Figure 154938DEST_PATH_IMAGE016
respectively representing the weight matrix of the decoder before and after each iteration of updating,
Figure 626370DEST_PATH_IMAGE017
which represents the zero norm of the vector,
Figure 618597DEST_PATH_IMAGE018
an exponential operation representing a vector representing a natural exponential function value raised to a power of a feature value at each position in the vector,
Figure 500971DEST_PATH_IMAGE005
it is shown that the addition by position,
Figure 211438DEST_PATH_IMAGE019
which represents a subtraction by position, is meant,
Figure 220983DEST_PATH_IMAGE012
representing a matrix multiplication.
10. An automatic dumping control system based on material level detection, characterized by comprising:
a level value acquisition unit for acquiring a plurality of level values acquired by a plurality of radar level gauges arranged in an area array;
the material level value arrangement unit is used for arranging the material level values into a material level input matrix according to the area array form;
the material level value space attention unit is used for enabling the material level input matrix to pass through a trained first convolution neural network model using a space attention mechanism so as to obtain a material level characteristic matrix;
the signal intensity value acquisition unit is used for acquiring the received signal intensity values of the plurality of radar level gauges when measuring the level values;
the signal intensity arrangement unit is used for arranging the received signal intensity values of the plurality of radar level gauges in the form of the area array as a signal intensity input matrix when measuring the level values;
the signal intensity space attention unit is used for enabling the signal intensity input matrix to pass through a trained second convolutional neural network model using a space attention mechanism so as to obtain a signal intensity characteristic matrix;
the decoding characteristic matrix generating unit is used for fusing the signal intensity characteristic matrix and the material level characteristic matrix to obtain a decoding characteristic matrix;
the decoding regression unit is used for performing decoding regression on the decoding characteristic matrix through a trained decoder to obtain a decoding value for representing the level correction measured value; and
and the comparison unit is used for determining whether to carry out dumping or not based on the decoding value.
CN202211188584.0A 2022-09-28 2022-09-28 Automatic soil discharging control method and system based on material level detection Active CN115571656B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202211188584.0A CN115571656B (en) 2022-09-28 2022-09-28 Automatic soil discharging control method and system based on material level detection

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202211188584.0A CN115571656B (en) 2022-09-28 2022-09-28 Automatic soil discharging control method and system based on material level detection

Publications (2)

Publication Number Publication Date
CN115571656A true CN115571656A (en) 2023-01-06
CN115571656B CN115571656B (en) 2023-06-02

Family

ID=84582420

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202211188584.0A Active CN115571656B (en) 2022-09-28 2022-09-28 Automatic soil discharging control method and system based on material level detection

Country Status (1)

Country Link
CN (1) CN115571656B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2024198266A1 (en) * 2023-03-27 2024-10-03 福建天甫电子材料有限公司 Production management control system for electronic-grade ammonium fluoride production

Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2004251001A (en) * 2003-02-20 2004-09-09 Shimizu Corp Earth discharging mechanism and earth discharging method for tunnel excavator
US20090091615A1 (en) * 2007-10-05 2009-04-09 Kar-Han Tan Optimization Strategies for GPU View Projection Matrix Implementation
CN101492750A (en) * 2008-12-30 2009-07-29 北京科技大学 High furnace burden face measurement and control system based on industrial phased array radar
CN106352951A (en) * 2016-09-29 2017-01-25 上海沃纳机电设备有限公司 Multi-frequency radar level gauge and material level measuring method based on same
CN111784777A (en) * 2020-07-09 2020-10-16 千顺智能(珠海)有限公司 SMT material quantity statistical method and statistical system based on convolutional neural network
CN112147589A (en) * 2020-08-18 2020-12-29 桂林电子科技大学 Frequency diversity array radar target positioning method based on convolutional neural network
CN112634292A (en) * 2021-01-06 2021-04-09 烟台大学 Asphalt pavement crack image segmentation method based on deep convolutional neural network
CN113627283A (en) * 2021-07-23 2021-11-09 中冶南方工程技术有限公司 Material level measuring method based on radar echo signals, terminal equipment and storage medium
CN114590605A (en) * 2022-03-15 2022-06-07 内蒙古京宁热电有限责任公司 Intelligent unmanned control system for bucket-wheel stacker reclaimer

Patent Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2004251001A (en) * 2003-02-20 2004-09-09 Shimizu Corp Earth discharging mechanism and earth discharging method for tunnel excavator
US20090091615A1 (en) * 2007-10-05 2009-04-09 Kar-Han Tan Optimization Strategies for GPU View Projection Matrix Implementation
CN101492750A (en) * 2008-12-30 2009-07-29 北京科技大学 High furnace burden face measurement and control system based on industrial phased array radar
CN106352951A (en) * 2016-09-29 2017-01-25 上海沃纳机电设备有限公司 Multi-frequency radar level gauge and material level measuring method based on same
CN111784777A (en) * 2020-07-09 2020-10-16 千顺智能(珠海)有限公司 SMT material quantity statistical method and statistical system based on convolutional neural network
CN112147589A (en) * 2020-08-18 2020-12-29 桂林电子科技大学 Frequency diversity array radar target positioning method based on convolutional neural network
CN112634292A (en) * 2021-01-06 2021-04-09 烟台大学 Asphalt pavement crack image segmentation method based on deep convolutional neural network
CN113627283A (en) * 2021-07-23 2021-11-09 中冶南方工程技术有限公司 Material level measuring method based on radar echo signals, terminal equipment and storage medium
CN114590605A (en) * 2022-03-15 2022-06-07 内蒙古京宁热电有限责任公司 Intelligent unmanned control system for bucket-wheel stacker reclaimer

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2024198266A1 (en) * 2023-03-27 2024-10-03 福建天甫电子材料有限公司 Production management control system for electronic-grade ammonium fluoride production

Also Published As

Publication number Publication date
CN115571656B (en) 2023-06-02

Similar Documents

Publication Publication Date Title
DK165026B (en) Method for detection of one or more distant objects
US7030814B2 (en) System and method to estimate the location of a receiver in a multi-path environment
US5923282A (en) Radar system
Nygren Terrain navigation for underwater vehicles
CA2454748A1 (en) Method for evaluating the measurement signals of a propagation-time based measurement device
CA1310389C (en) System for detecting underground objects
US10180342B2 (en) Level finding using multiple search steps
CN107705359B (en) Channel modeling method and device using three-dimensional visual reconstruction technology
CN115571656A (en) Automatic dumping control method and system based on material level detection
KR101897763B1 (en) Method and apparatus for distance measurement using radar
CN110426717B (en) Cooperative positioning method and system, positioning device and storage medium
JP2020181571A (en) Chipless RFID Decoding System and Method
CN107102323A (en) For the improved method and system of fmcw radar altimeter system altitude Measurement Resolution
RU2576341C2 (en) Phase-based tracking
CN106597395A (en) Real-time pulse azimuth detecting method and radar equipment
JP4318189B2 (en) Non-destructive exploration device, method and program
KR102195232B1 (en) Wireless power transfer system and method of beamforming weight estimating
CN110991507A (en) Road underground cavity identification method, device and system based on classifier
JP2020173160A (en) Tsunami height and tsunami arrival time prediction system
CN117351321A (en) Single-stage lightweight subway lining cavity recognition method and related equipment
EP0956503B1 (en) Apparatus for determining properties of an electrically conductive object
CN111007464B (en) Road underground cavity identification method, device and system based on optimal weighting
US11693105B2 (en) Electromagnetic wave field data processing method and apparatus, and medium
CN113970774A (en) Ambiguity fixing method and device of navigation system
KR20210054769A (en) Radar signal prcessing apparatus and radar signal processing method

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant